Global coordination under monotone sampling

We study coordination in multi-population environments where agents revise actions after observing finite independent samples of others’ behavior and applying deterministic monotone decision rules. The framework covers a broad class of commonly studied learning dynamics, and it allows heterogeneity across populations. Except for the knife-edge case in which every population simply copies one sampled action, coordination is globally stable: almost every trajectory converges to a coordinated state in which everyone chooses the same action. The proof connects monotone sampling rules to reliability functions of coherent systems. We show that relaxing independence, monotonicity, or within-population homogeneity can generate stable miscoordination.

(joint with Srinivas Arigapudi)

Link to the paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6681838
Yuval Heller's homepage: https://sites.google.com/site/yuval26